Large Language Models in Clinical Neurology: A Systematic Review

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Medicine
• Neurology

Background:

• Large language models (LLMs) are being explored for clinical neurology applications.
• Uncertainty exists regarding their real-world utility, safety, and optimal implementation.
• A systematic review was conducted to address these knowledge gaps.

Purpose of the Study:

• To systematically review and characterize current LLM applications in clinical neurology.
• To evaluate the quality of evidence for LLM use in neurology.
• To identify knowledge gaps and future directions for LLM implementation in neurology.

Main Methods:

• Systematic literature search following PRISMA guidelines across multiple databases (PubMed, Embase, Scopus, Web of Science, CENTRAL) from January 2022 to February 2026.
• Inclusion of peer-reviewed studies evaluating LLMs for clinically relevant neurology tasks using text or multimodal inputs.
• Data extraction and risk of bias assessment using QUADS-AI by two independent reviewers; narrative synthesis of evidence.

Main Results:

• Thirty-six studies (2023-2026) across 8 neurology subspecialties were included; most were simulation or retrospective analyses.
• LLMs demonstrated high performance in constrained tasks like diagnostic classification (AUC 0.75-0.94) and information extraction (F1 0.89-0.90).
• Safety concerns include hallucinations, overconfident recommendations, and high risk of bias across all studies; lower accuracy in open-ended tasks.

Conclusions:

• LLMs show potential for specific neurology workflows, but current evidence is preliminary and limited by heterogeneity and bias.
• Clinical translation necessitates retrieval-augmented generation (RAG) and agentic architectures for multi-step task planning, verification, and auditable outputs.
• Clinician oversight and prospective validation are crucial for safe and effective integration of LLMs in clinical neurology.